Intel Stresses New Selling Points In Server Chip

Intel once again turned the crank Tuesday on a new generation of chips for server systems. But there were some differences in the messaging this time, a sign of how the business is changing.

The company used to stress how much faster each new electronic brain worked, measured by megahertz or gigahertz. Then the race shifted to how many such processor cores could be put on a single chip. But both approaches reached a point of diminishing returns, because of factors such as excess power consumption and programming difficulties.

Meanwhile, computer makers and their customers are demanding new kinds of features amid the rise of outsourced service providers–the trend known as cloud computing–and as more computing jobs are packed onto each server by means of the technology called virtualization.

Intel says its new Xeon chip, dubbed the E5-2600, has special circuitry and internal instructions to speed up specific kinds of computing chores, and boost the overall capability of a computer to get work done. The sum of all the improvements, the company said during a briefing in San Francisco, translate into up to 80% greater performance than prior Xeons.

In one key change, Intel engineers took what had been a separate communications chip and placed that circuitry on the same piece of silicon as the microprocessor, reducing delays in moving data from the outside world and from memory chips in a system. It added special instructions that work with hardware improvements to speed up what scientists call floating-point calculations, doubling performance in some cases, the company said.

In another example, Intel also added new instructions that encrypt data much more quickly than earlier hardware–a capability long sought by security conscious customers.

Diane Bryant, an Intel vice president recently appointed as general manager of the group that developed the chip, said some 400 pieces of hardware are being designed with it. The customers include well-known server makers such IBM, Dell, Hewlett-Packard and Silicon Graphics.

But Intel is also claiming traction in using Xeon chips as the calculating engine for more specialized devices that handle communications and data storage.

“It is not just an outstanding server solution,” Bryant said. “It is a storage solution and a networking solution.”

Matt Eastwood, an analyst at IDC, stressed that the biggest target are companies that buy big volumes of servers that have sockets for one or two microprocessor chips. Such customers include makers of the largest machines for scientific tasks, called supercomputers, as well as companies like Google that buy thousands of servers for repetitive Web tasks.

Intel, in another departure from past practice, last fall released some early shipments of the new Xeon chip for those two classes of customers. The tactic, among other things, helped the company respond to competition from the latest version of the Opteron chip from AMD, Eastwood said.

The new chip comes in models with four, six or eight processor cores, up to 20 megabytes of built-in memory and clock speeds up to 3.6 gigahertz. List prices range from $198 to $2,050.

While it wasn’t the focus of the Intel press conference, one attendee kicked off a question-and-answer session by asking Bryant about a recent deal by AMD to buy SeaMicro, a company that up to now has been selling what the industry calls “micro servers” based on Intel chips.

While another Intel executive declined comment on the matter last week, Bryant said her company was offered a chance to buy SeaMicro or its technology, too. “There are very few people they didn’t shop their solution to,” she said. “We were not impressed and we declined.”